You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
notepad-plus-plus/lexilla/lexers/LexConf.cxx

194 lines
5.6 KiB

// Scintilla source code edit control
/** @file LexConf.cxx
** Lexer for Apache Configuration Files.
**
** First working version contributed by Ahmad Zawawi <ahmad.zawawi@gmail.com> on October 28, 2000.
** i created this lexer because i needed something pretty when dealing
** when Apache Configuration files...
**/
// Copyright 1998-2001 by Neil Hodgson <neilh@scintilla.org>
// The License.txt file describes the conditions under which this software may be distributed.
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
#include <stdarg.h>
#include <assert.h>
#include <ctype.h>
#include <string>
#include <string_view>
#include "ILexer.h"
#include "Scintilla.h"
#include "SciLexer.h"
#include "WordList.h"
#include "LexAccessor.h"
#include "Accessor.h"
#include "StyleContext.h"
#include "CharacterSet.h"
#include "LexerModule.h"
using namespace Lexilla;
static void ColouriseConfDoc(Sci_PositionU startPos, Sci_Position length, int, WordList *keywordLists[], Accessor &styler)
{
int state = SCE_CONF_DEFAULT;
char chNext = styler[startPos];
Sci_Position lengthDoc = startPos + length;
// create a buffer large enough to take the largest chunk...
char *buffer = new char[length+1];
Sci_Position bufferCount = 0;
// this assumes that we have 2 keyword list in conf.properties
WordList &directives = *keywordLists[0];
WordList &params = *keywordLists[1];
// go through all provided text segment
// using the hand-written state machine shown below
styler.StartAt(startPos);
styler.StartSegment(startPos);
for (Sci_Position i = startPos; i < lengthDoc; i++) {
char ch = chNext;
chNext = styler.SafeGetCharAt(i + 1);
if (styler.IsLeadByte(ch)) {
chNext = styler.SafeGetCharAt(i + 2);
i++;
continue;
}
switch(state) {
case SCE_CONF_DEFAULT:
if( ch == '\n' || ch == '\r' || ch == '\t' || ch == ' ') {
// whitespace is simply ignored here...
styler.ColourTo(i,SCE_CONF_DEFAULT);
break;
} else if( ch == '#' ) {
// signals the start of a comment...
state = SCE_CONF_COMMENT;
styler.ColourTo(i,SCE_CONF_COMMENT);
} else if( ch == '.' /*|| ch == '/'*/) {
// signals the start of a file...
state = SCE_CONF_EXTENSION;
styler.ColourTo(i,SCE_CONF_EXTENSION);
} else if( ch == '"') {
state = SCE_CONF_STRING;
styler.ColourTo(i,SCE_CONF_STRING);
} else if( IsASCII(ch) && ispunct(ch) ) {
// signals an operator...
// no state jump necessary for this
// simple case...
styler.ColourTo(i,SCE_CONF_OPERATOR);
} else if( IsASCII(ch) && isalpha(ch) ) {
// signals the start of an identifier
bufferCount = 0;
buffer[bufferCount++] = static_cast<char>(tolower(ch));
state = SCE_CONF_IDENTIFIER;
} else if( IsASCII(ch) && isdigit(ch) ) {
// signals the start of a number
bufferCount = 0;
buffer[bufferCount++] = ch;
//styler.ColourTo(i,SCE_CONF_NUMBER);
state = SCE_CONF_NUMBER;
} else {
// style it the default style..
styler.ColourTo(i,SCE_CONF_DEFAULT);
}
break;
case SCE_CONF_COMMENT:
// if we find a newline here,
// we simply go to default state
// else continue to work on it...
if( ch == '\n' || ch == '\r' ) {
state = SCE_CONF_DEFAULT;
} else {
styler.ColourTo(i,SCE_CONF_COMMENT);
}
break;
case SCE_CONF_EXTENSION:
// if we find a non-alphanumeric char,
// we simply go to default state
// else we're still dealing with an extension...
if( (IsASCII(ch) && isalnum(ch)) || (ch == '_') ||
(ch == '-') || (ch == '$') ||
(ch == '/') || (ch == '.') || (ch == '*') )
{
styler.ColourTo(i,SCE_CONF_EXTENSION);
} else {
state = SCE_CONF_DEFAULT;
chNext = styler[i--];
}
break;
case SCE_CONF_STRING:
// if we find the end of a string char, we simply go to default state
// else we're still dealing with an string...
if( (ch == '"' && styler.SafeGetCharAt(i-1)!='\\') || (ch == '\n') || (ch == '\r') ) {
state = SCE_CONF_DEFAULT;
}
styler.ColourTo(i,SCE_CONF_STRING);
break;
case SCE_CONF_IDENTIFIER:
// stay in CONF_IDENTIFIER state until we find a non-alphanumeric
if( (IsASCII(ch) && isalnum(ch)) || (ch == '_') || (ch == '-') || (ch == '/') || (ch == '$') || (ch == '.') || (ch == '*')) {
buffer[bufferCount++] = static_cast<char>(tolower(ch));
} else {
state = SCE_CONF_DEFAULT;
buffer[bufferCount] = '\0';
// check if the buffer contains a keyword, and highlight it if it is a keyword...
if(directives.InList(buffer)) {
styler.ColourTo(i-1,SCE_CONF_DIRECTIVE );
} else if(params.InList(buffer)) {
styler.ColourTo(i-1,SCE_CONF_PARAMETER );
} else if(strchr(buffer,'/') || strchr(buffer,'.')) {
styler.ColourTo(i-1,SCE_CONF_EXTENSION);
} else {
styler.ColourTo(i-1,SCE_CONF_DEFAULT);
}
// push back the faulty character
chNext = styler[i--];
}
break;
case SCE_CONF_NUMBER:
// stay in CONF_NUMBER state until we find a non-numeric
if( (IsASCII(ch) && isdigit(ch)) || ch == '.') {
buffer[bufferCount++] = ch;
} else {
state = SCE_CONF_DEFAULT;
buffer[bufferCount] = '\0';
// Colourize here...
if( strchr(buffer,'.') ) {
// it is an IP address...
styler.ColourTo(i-1,SCE_CONF_IP);
} else {
// normal number
styler.ColourTo(i-1,SCE_CONF_NUMBER);
}
// push back a character
chNext = styler[i--];
}
break;
}
}
delete []buffer;
}
static const char * const confWordListDesc[] = {
"Directives",
"Parameters",
0
};
extern const LexerModule lmConf(SCLEX_CONF, ColouriseConfDoc, "conf", 0, confWordListDesc);